Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Embedding Non-Distortive Cancelable Face Template Generation

Dmytro Zakharov, Oleksandr Kuznetsov, Emanuele Frontoni, Natalia Kryvinska

TL;DR

Biometric systems face privacy risks when storing raw biometric data. This work proposes non-distortive cancelable biometrics, distorting face images to appear different to humans while preserving identity in a learned embedding space via an Embedding Model $E$ and a Distortion Generator $G$. The approach optimizes a composite loss that balances perceptual distortion $d_{img}$ and embedding stability $d_{emb}$, implemented with a Trainer Network and evaluated on MNIST and LFW using a UNet-based generator and either pretrained FaceNet or a custom embedding network. Results show maintained or improved accuracy under distortion while delivering strong privacy safeguards, supporting practical deployment of privacy-aware biometric authentication.

Abstract

Biometric authentication systems are crucial for security, but developing them involves various complexities, including privacy, security, and achieving high accuracy without directly storing pure biometric data in storage. We introduce an innovative image distortion technique that makes facial images unrecognizable to the eye but still identifiable by any custom embedding neural network model. Using the proposed approach, we test the reliability of biometric recognition networks by determining the maximum image distortion that does not change the predicted identity. Through experiments on MNIST and LFW datasets, we assess its effectiveness and compare it based on the traditional comparison metrics.

Embedding Non-Distortive Cancelable Face Template Generation

TL;DR

Biometric systems face privacy risks when storing raw biometric data. This work proposes non-distortive cancelable biometrics, distorting face images to appear different to humans while preserving identity in a learned embedding space via an Embedding Model and a Distortion Generator . The approach optimizes a composite loss that balances perceptual distortion and embedding stability , implemented with a Trainer Network and evaluated on MNIST and LFW using a UNet-based generator and either pretrained FaceNet or a custom embedding network. Results show maintained or improved accuracy under distortion while delivering strong privacy safeguards, supporting practical deployment of privacy-aware biometric authentication.

Abstract

Biometric authentication systems are crucial for security, but developing them involves various complexities, including privacy, security, and achieving high accuracy without directly storing pure biometric data in storage. We introduce an innovative image distortion technique that makes facial images unrecognizable to the eye but still identifiable by any custom embedding neural network model. Using the proposed approach, we test the reliability of biometric recognition networks by determining the maximum image distortion that does not change the predicted identity. Through experiments on MNIST and LFW datasets, we assess its effectiveness and compare it based on the traditional comparison metrics.
Paper Structure (16 sections, 9 equations, 4 figures, 1 table, 2 algorithms)

This paper contains 16 sections, 9 equations, 4 figures, 1 table, 2 algorithms.

Table of Contents

  1. Introduction
  2. Background
  3. Methodology
  4. Embedding Model
  5. Generator Model
  6. Loss Function
  7. Trainer Network Architecture
  8. Implementation
  9. Embedding model
  10. Generator model
  11. Results
  12. Hamming Distance Analysis
  13. Test Recognition System
  14. Performance Metrics Evaluation
  15. Discussion and Comparison of Results
  16. ...and 1 more sections

Figures (4)

  • Figure 1: An example of using our proposed image distortion technique on images from MNISTmnist and LFWlfw datasets. While authentic and generated images significantly differ, the feature vectors of both images in pairs are relatively close.
  • Figure 2: Trainer Network architecture
  • Figure 3: Embedding model architecture.
  • Figure 4: Histogram of Hamming distances for three cases: "Real vs Real photos", "Real vs Generated photos", and "Generated vs Generated photos" and two different datasets: (a)LFW, (b)MNIST.

Theorems & Definitions (2)

  • definition thmcounterdefinition
  • definition thmcounterdefinition